The new economics of AI aren't just about compute power; they're about token spend. And for ambitious founders in their 20s and 30s, the playbook for this new reality is sharply divided. Mike Mignano, a new General Partner at USV, laid out a contrarian view on 20VC with Harry Stebbings: startups should aggressively maximize their token spend on frontier models, while large enterprises have no choice but to constrain it.

Key Takeaways

  • Startups must aggressively maximize token spend on frontier AI models to gain a competitive advantage, a strategy large enterprises cannot afford.
  • Large incumbents like Salesforce face an existential challenge: unlimited AI token spend across their vast employee base would wreck their business fundamentals.
  • 80% of enterprise non-coding tasks can be handled by less expensive, non-frontier models, freeing up frontier models for high-value coding work.
  • The "routing layer" is emerging as a critical opportunity, allowing companies to dynamically select the most efficient AI model based on task capability and cost.
  • New monetization models, like a "bounty system" where the router is paid for optimal model choice, could arise in this routing layer.

Your AI Token Strategy: Maximize or Constrain?

Forget what you heard about cost-cutting in AI. Mike Mignano argues that your approach to AI token spend depends entirely on your company's scale. If you're a startup, your goal isn't to be cheap; it's to be better and faster. “As a startup you need every advantage you can get right now,” Mignano said. “If I were the CEO of a startup right now I would actually still be pounding the table to maximize token spend on the on the right things.”

This isn't reckless spending; it's a calculated bet on superior capability. Frontier models offer an edge, and for a lean team, that edge can translate directly into product breakthroughs or operational efficiency that outpaces larger, slower rivals. Your burn rate might tick up slightly, but your chances of finding product-market fit or unlocking new features dramatically improve. This is a crucial distinction: spend to win, not just to save.

Large enterprises, on the other hand, face a terrifying math problem. Imagine Salesforce, with its thousands of employees, each consuming frontier model tokens without limits. “If every employee is just spending like crazy on tokens, the fundamentals of these businesses are going to be in trouble,” Mignano warned. “They're just too big, right? there's too many employees at a Salesforce or I don't know a Microsoft such that every employee can just have an unlimited token spend budget.” For these giants, reigning in token usage isn't optional; it's a matter of survival.

The New Opportunity: The AI Routing Layer

This fundamental divergence in token strategy creates a massive opportunity: the routing layer. As enterprises grapple with the need to constrain costs while still using AI, they won't simply cut off access. Instead, they'll get smart about which model gets used for which task. Mignano believes “routing is interesting and important right now... as enterprises are trying to optimize their token spend, they want to make sure that the model they're using is the right model for the job. Not only in terms of its capability and what it can get done, but its cost.”

Here's why this matters for you: most tasks don't need a frontier model. Mignano noted, “80% of non-coding tasks in the enterprise can be done with models that are not at the frontier.” The routing layer intelligently directs queries to the cheapest, most capable model for a given job. For a simple text summary, why pay for GPT-4 when a smaller, faster, cheaper model can do it? But for complex code generation, you still want the best. This creates a market for companies that can build or provide this intelligent routing.

One intriguing idea Mignano heard was a "bounty model" for routers. “The routing layer gets rewarded for choosing the right model,” he explained. “So if you choose the most efficient model or the best model for a certain task, that's when they collect a fee.” This moves beyond simple margins, turning routing into a performance-based monetization engine.

What to Do With This

Tomorrow, pull up your current AI API spend. If you're a startup, identify where maximizing spend on frontier models could accelerate your core product development or create a significant competitive edge, and advocate for that budget internally. Simultaneously, for any commodity or internal tasks, begin actively exploring and integrating smaller, cheaper models. If your team's AI usage is scaling, start designing or researching a "routing layer" architecture now, considering how you'd dynamically choose models based on both cost and specific task capability. The future of AI profitability lies in this intelligent model orchestration, not just raw compute.